Electrical impedance tomography: methods and applications

Abstract

Electrical impedance tomography (EIT) is an imaging technique for mapping the internal conductivity distribution of an object by taking voltage measurements from electrodes attached to the surface of the object, while an electrical current is injected to these boundary electrodes. EIT has been researched in many different application areas in the world as a simpler, cheaper and safer alternative to many other tomography techniques, while provider very useful and often unique functional information. The main aim of this PhD study is to extend the use of EIT by further improvements of our understanding of the EIT data and image analysis and its challenges. In each case the challenges are highlighted and some solutions are proposed.

EIT has great applications in area of industrial processes. We highlight a challenges associated with the EIT to simultaneously reconstruct the permittivity and conductivity, in particular, when there is a low contrast in permittivity values of samples but in high contrast with background. A good case study to highlight this challenge is a water dominate oil and gas ow with important application in process industry. The thesis is proposing a dual-modality EIT with transmission mode travel time ultrasound tomography for such an application.

Another potential area with the EIT is the use of EIT for artificial skin for robotics application. This is done using a soft material such as fabric and the skin can be developed by EIT with sensing the change in conductivity while pressure applied to the skin. We have identified two main issues with the application, one the need to extend the functionality of the skin to be dynamical. This will enable the EIT based skin to work as an interface allowing social interaction with the robot. Second is a well established issue in medical EIT, which also exists in robotics EIT and that is the movement of electrodes which corrupted the EIT image. We have developed a spatially correlated total variation imaging algorithm so that the robot skin using EIT could work as dynamical imaging sensor allowing for interactive skin. The interaction of the EIT based skin through pressure sensing can be done like a movie rather than individual images, which resembles the human skin interaction. The movement of electrodes and electromechnaicl interpretation of pressure via EIT image are both very di cult problems to model and interpreter. For these issue we implemented a convolutional neural network deep learning algorithm. The imaging results shows very good performance of both spatially correlated TV algorithm together with the deep learning approach.

The working ow of this dissertation can be explained as the following sections. Firstly, the basic background of EIT, its applications and mathematical theories including the forward problem, inverse problem have been reviewed. Secondly, a complex impedance image reconstruction is developed. The complex EIT which is determining conductivity and permittivity distribution at the same time using the real and imaginary part of the voltage measurements are presented. A complex valued forward model, Jacobian matrix, inverse solution and related simulation studies are developed as well, the results indicated there are still challenging in reconstructing both parameters simultaneously. And then, a novel EIT combined with ultrasound transmission mode tomography (UTT) dual-modality for three phase material image is developed. Identification of three phase oil/gas/water in water dominated situation should be possible via complex EIT, but practicality this is challenging. Therefore, the EIT/UTT dual modality imaging can be deployed for such application, where EIT is used to identify non-conductive phase which either oil or gas phase and hence UTT is used to identify air phase. Both simulation and experimental studies are implemented and a image fusion method is proposed for producing three-phase images. Finally, a conductive fabric based EIT dynamical sensing system integrated with deep learning for improving image quality is proposed, a movie like denoised experimental results are presented using a spatiotemporal total variation algorithm and a convolutional neural network training. The deep learning method helps overcoming the imaging artefacts due to electrode movement which is a main issue in fabric based EIT.

Date of Award	24 Jun 2020
Original language	English
Awarding Institution	University of Bath
Supervisor	Manuchehr Soleimani (Supervisor) & Biagio Forte (Supervisor)